Latent class analysis to identify subphenotypes predicting pediatric splenic pseudoaneurysm following blunt spleen injuries: A post-hoc analysis

This study is a post hoc analysis of a multicenter, retrospective observational study conducted by the Splenic and Hepatic Injury in Pediatric Patients (SHIPPs) study group [1]. The SHIPPs registry compiled pre- and in-hospital data on pediatric patients aged ≤16 years who sustained blunt liver and/or splenic injuries and were admitted to 83 hospitals across Japan between January 2008 and December 2019. Data were entered into a web-based system by medical personnel at each participating institution. Outcome assessors were not blinded to clinical information.

The original SHIPPs study and its subsequent analyses were approved by the Ethics Committee of Jichi Medical University Saitama Medical Center (approval number: S20-112). As this investigation is a secondary analysis of the SHIPPs registry, no additional ethical approval was required. Informed consent was waived, as the study involved no interventions deviating from standard clinical practice; however, an opt-out option was made available on the institution’s website. The data were accessed for research purposes on June 6, 2024. No personally identifiable information was available to the investigators during or after data collection.

This study was conducted in accordance with the STrengthening the Reporting of OBservational Studies in Epidemiology (STROBE) guidelines and the principles outlined in the Declaration of Helsinki and its subsequent amendments (Table S1 in Supplementary Online File 1) [10].

The inclusion criteria for the SHIPPs study were as follows: (1) age ≤16 years, (2) transportation by ambulance, (3) diagnosis of splenic and/or liver injury upon hospital arrival, and (4) an Abbreviated Injury Scale (AIS) score ≥1 related to the liver or spleen [1]. The exclusion criteria were: (1) cardiopulmonary arrest on arrival, (2) any injury with an AIS score of 6, (3) duplicate enrollment due to interhospital transfers from both referring and receiving institutions, and (4) refusal of treatment or request for treatment limitation by the parent or guardian due to severe head injury [1].

For the present analysis, additional exclusion criteria were applied: (1) isolated liver injury or combined liver and splenic injuries, (2) cases requiring surgical intervention, (3) pseudoaneurysm formation identified on the day of admission, (4) missing data regarding delayed splenic pseudoaneurysm formation, and (5) transfer to another hospital within 5 days of admission without subsequent follow-up information. In the SHIPPs study, the study cohort was divided into four groups based on hemostatic intervention within 48 hours of admission: NOM (observed), NOM with IVR, operative management (OM), and combined IVR/OM [1]. In contrast, in the current study, we included only cases that fell under NOM (observed), excluding those who underwent surgery or required emergency IVR due to the presence of a splenic pseudoaneurysm at the time of admission.

The following data were collected: age, sex, weight, past medical history (hematologic disease, neuropsychiatric disorder, chromosomal abnormalities, congenital anomaly, asthma, others), cause of injury (fall, fall down, sport, bicycle, motor vehicle crash, struck by vehicle, abuse, assault, others), vital signs on arrival (heart rate [HR], blood pressure [BP]), presence or absence of shock on arrival, results of blood tests conducted on arrival (hemoglobin [Hb] level, platelet [Plt] count, prothrombin time-international normalized ratio [PT-INR]), initial CT findings (capsular tear of the spleen [none, yes, unknown], presence or absence of contrast extravasation (CE) in the spleen, presence or absence of intra-abdominal bleeding, concomitant injury to other organs), organ injury scaling (OIS) 2018 grade of the spleen on arrival, injury severity score (ISS), blood transfusion, cryoprecipitate transfusion, fibrinogen transfusion, infusion of tranexamic acid, the method of IVR (angiography only, embolization with Gelatin Sponge, embolization with coil, others, unknown), duration of hospitalization, 30-day mortality, hospital mortality, and formation of delayed splenic pseudoaneurysm [11,12].

Shock was defined according to the Pediatric Advanced Life Support (PALS) guidelines of the American Heart Association [13]. For transfusion, distinctions were not made between the types of red blood cells (RBCs), fresh frozen plasma (FFP), or Plt.

The outcome measure was delayed splenic pseudoaneurysm formation. This outcome measure was defined as the formation of a splenic pseudoaneurysm, which was detected on or after the second day after injury (but undetectable on CT scan on admission). The determination of whether delayed splenic pseudoaneurysm had formed was left to the discretion of the physicians at each facility.

We selected the following clinically relevant variables for latent class analysis (LCA): age, sex, weight, past medical history, mechanism of injury, presence or absence of shock on arrival, blood test results at admission, initial CT findings, splenic injury grade based on the 2018 Organ Injury Scale (OIS), Injury Severity Score (ISS), blood transfusion during hospitalization, administration of cryoprecipitate, fibrinogen transfusion, tranexamic acid infusion, and the type of IVR procedure performed. Multiple imputation of explanatory variables was conducted using the Markov chain Monte Carlo method. Given that the maximum proportion of missing data was 24%, a total of 24 imputed datasets were generated [14]. Although the exact timing of transfusions was not recorded, we assumed they were administered during the acute phase immediately after injury—prior to the potential formation of delayed splenic pseudoaneurysms.

We initially explored models with 2 to 5 latent classes and determined the optimal number of clinically meaningful sub-phenotypes based on both the Bayesian Information Criterion (BIC) and clinical interpretability. After identifying the optimal class structure, the discriminative capacity of each variable was evaluated using the maximum integrated complete-data likelihood criterion. Higher variable indices indicated stronger associations between the variables and class membership.

After sub-phenotypes were established, continuous variables were summarized as medians and interquartile ranges (IQRs), while categorical variables were reported as absolute numbers and percentages. Differences in variables among the latent classes were assessed using Pearson’s chi-squared test or the Kruskal–Wallis rank sum test, as appropriate. Logistic regression analysis was subsequently performed to examine the association between each sub-phenotype and the occurrence of delayed splenic pseudoaneurysms, expressed as odds ratios (ORs) with corresponding 95% confidence intervals (CIs). No covariate adjustment was performed, as the latent classes were conceptualized as unobserved homogeneous subgroups based on patient characteristics, and thus covariates were not considered to confound the relationship between latent classes and outcomes [15].

All analyses were performed using the R package VarSelLCM and R statistical software (version 4.1.3; The R Project for Statistical Computing, Vienna, Austria). A two-sided p-value of <0.05 was considered statistically significant.

Of the 1,441 patients included in the registry of pediatric hepatic and splenic injuries, we analyzed the data of 434 patients who met the inclusion criteria for the current study (Figure 1).

Fig. 1.

Flowchart depicting the screening and enrollment process in this study

Three Sub-phenotypes were identified after LCA (Table S2 in Supplementary Online File 2). Patient characteristics and pre- and post-hospital information for each latent class are listed in Table 1, and missing values are shown in Table S3 in Supplementary Online file 2.

The factor with the highest discriminative power was PT-INR (Figure 2). The following values are all listed in the order of Sub-phenotype 1, Sub-phenotype 2, and Sub-phenotype 3. The median (IQR) values of PT-INR were as follows: 1.27 (1.18–1.56), 1.15 (1.08–1.24), and 1.08 (1.02–1.14); and plt on arrival were as follows: 22.8 (15.7–28.1) ×10³/μL, 23.9 (20.8–28.7) ×10³/μL, and 28.9 (23.1–33.3) ×10³/μL (Table 1, and Figure S1-a and S1-b in Supplementary Online File 2). Among the other variables, shock was observed in six patients (18.8%), eight patients (4.2%), and five patients (2.4%); CE in the spleen of initial CT was observed in 22 patients (68.8%), 49 patients (25.7%), and 22 patients (10.4%); OIS grade 4 or 5 for the spleen was observed in 19 patients (59.4%), 44 patients (23.0%), and 31 patients (14.7%); the median (IQR) values of ISS were as follows: 31.5 (21.5–42.3), 9 (5.5–17), and 9 (5–14); and blood transfusion was performed in 25 patients (78.1%), 25 patients (13.1%), and 19 patients (9.0%) (Table 1, and Figure S1-c–S1-e in Supplementary Online File 2). And the variables discussed above, although not ranked among those with the highest discriminative power, are presented in Table S4 in Supplementary Online File 2. Furthermore, the variables with statistically significant differences among the LCA-derived sub-phenotypes are shown in Table S5 in Supplementary Online File 2.

Fig. 2.

The five factors showing high discriminative power in LCA. Abbreviations: ISS, Injury Severity Scale score; LCA, latent class analysis; Plt, platelet count; PT-INR, prothrombin time-international normalized ratio.

In total, delayed splenic pseudoaneurysm formation was observed in 46 patients (10.6%), including seven patients (21.9%) in Sub-phenotype 1, 25 patients (13.1%) in Sub-phenotype 2, and 14 patients (6.6%) in Sub-phenotype 3 (p = 0.01; Table 2 and Figure 3).

Fig. 3.

Formation of delayed splenic pseudoaneurysm in each Sub-phenotype

Univariate logistic regression analysis for delayed splenic pseudoaneurysm formation using Sub-phenotype 3 as the reference revealed an OR (95% CI) of 3.94 (1.45–10.7) in Sub-phenotype 1 and 2.12 (1.07–4.21) in Sub-phenotype 2 (Table 3).